<alyssa>
and I haven't tested full throttle to see if there'd be crashes
<alyssa>
I can try again
mattia013_ has joined #asahi-gpu
chrisl has joined #asahi-gpu
chrisl has quit [Ping timeout: 480 seconds]
mattia013_ has quit [Ping timeout: 480 seconds]
pb17 has quit [Ping timeout: 480 seconds]
pb17 has joined #asahi-gpu
chadmed has quit [Read error: Connection reset by peer]
chadmed has joined #asahi-gpu
chrisl has joined #asahi-gpu
chrisl has quit [Ping timeout: 480 seconds]
Larwive has joined #asahi-gpu
chrisl has joined #asahi-gpu
allmazz has joined #asahi-gpu
Larwive has quit [Ping timeout: 480 seconds]
chrisl has quit [Ping timeout: 480 seconds]
Larwive has joined #asahi-gpu
<alyssa>
jannau: been running vulkan CTS a bunch on m2 max 12 threads, indeed no crash
<alyssa>
this is really weird.
codingkoopa3218 has quit []
codingkoopa3218 has joined #asahi-gpu
<alyssa>
what changed? kernel or mesa? :/
<alyssa>
half inclined to bisect this so we know if/when this resurfaces agian
<alyssa>
^know what to do
chadmed has quit [Quit: Konversation terminated!]
chadmed has joined #asahi-gpu
chadmed has quit []
chadmed has joined #asahi-gpu
chadmed has quit []
chadmed has joined #asahi-gpu
Larwive has quit [Ping timeout: 480 seconds]
<jannau>
alyssa: maybe the our kernel submit "regression" made it unlikely to hit. maybe you fixed something while reworking the uapi
<jannau>
bisecting would be annoying over the uapi changes
<alyssa>
jannau: mm, right.
<alyssa>
I.. *think* I remember seeing crashes even with the new uapi..
<jannau>
while or after the changes?
<jannau>
are you testing with the stock fedora kernel or slef-built asahi-wip?
<alyssa>
stock 6.14.4-400.asahi.fc42.aarch64+16k
<alyssa>
I don't remember the last time I saw crashes though
<jannau>
ok, that at least confirms that's not the devicetree/initdata fixes
<alyssa>
:+1:
<alyssa>
that reminds me, we should probably submit the dt-bindings to the list before shipping new m1n1s since we don't want to do that dance twice
<jannau>
we could test whether the asahi/hk enable PR crashes the firmware. if it does bisecting would be easy
<alyssa>
i'll run that over lunch
<jannau>
was robher ok with stuffing 27kb + 6kb + 92kb (or so for the Ultras) of binary data into the dtb? I find that at least inconvenient since I occasionally look at the decompiled devicetree from /proc/device-tree
<jannau>
the alternative would be using reserved-memory for the initdata
Larwive has joined #asahi-gpu
enick_477 has quit [Quit: Bridge terminating on SIGTERM]
rhysmdnz has quit [Quit: Bridge terminating on SIGTERM]
Jamie has joined #asahi-gpu
Jamie is now known as Guest16789
rhysmdnz has joined #asahi-gpu
<alyssa>
I don't think I mentioned sizes. I didn't realize it was that big, thought it was a few kb's only
allmazz has quit [Quit: Lost terminal]
WindowPain has quit [Ping timeout: 480 seconds]
WindowPain has joined #asahi-gpu
DragonStar has joined #asahi-gpu
chrisl has joined #asahi-gpu
pb17 has quit [Ping timeout: 480 seconds]
VG9iaWFz has quit [Remote host closed the connection]
pb17 has joined #asahi-gpu
chrisl has quit [Ping timeout: 480 seconds]
DragonStar has quit [Ping timeout: 480 seconds]
DragonStar has joined #asahi-gpu
chadmed has quit [Quit: Konversation terminated!]
cylm has joined #asahi-gpu
chrisl has joined #asahi-gpu
karolherbst7 has joined #asahi-gpu
karolherbst has quit [Read error: Connection reset by peer]
<alyssa>
jannau: just hit a crash
<alyssa>
(on my current mesa wip branch)
<alyssa>
so I'd say the crashes have maybe become less frequent lately (possibly the submit perf regression), but it's definitely not gone
<alyssa>
i'm almost relieved tbh
chrisl has quit [Ping timeout: 480 seconds]
<jannau>
ack. little annoying that it makes testing the devicetree changes harder. I guess I could use old kernel and mesa
karolherbst7 has quit []
karolherbst has joined #asahi-gpu
DragonStar has quit [Read error: Connection timed out]
<alyssa>
jannau: was there ever a kernel/mesa that suffer firmware crashes on the big machines?
<alyssa>
(also, have we observed this crash ever on pre-g14x?)
<jannau>
it easily reproduced on g13d. I haven't tested that much on g13c
<alyssa>
hmm ok
<alyssa>
i mostly ask because, I only started running CTS on newer-than-g13g in september
<alyssa>
and i'm not sure there was ever a time when it was, rock solid since then
<jannau>
had to use --jobs 6 or less to get full CTS runs. I saw that before september both with GL and VK CTS
<alyssa>
right.. so not sure there was ever a working kernel/mesa
<jannau>
I think it was an issue from when I started running CTS on the studio
DragonStar has joined #asahi-gpu
<jannau>
the apple,ppm-ki "<30>" vs. "<30.0>" typo is there since asahi-6.1-1 (likely the beginning of GPU support)
<alyssa>
oops
DragonStar has quit [Ping timeout: 480 seconds]
DragonStar has joined #asahi-gpu
<jannau>
does the CTS cache shaders? the old CTS build works again after downgrading mesa to asahi-20250221
chrisl has joined #asahi-gpu
pb17 has quit [Ping timeout: 480 seconds]
pb17 has joined #asahi-gpu
DragonStar has quit [Ping timeout: 480 seconds]
chrisl has quit [Ping timeout: 480 seconds]
DragonStar has joined #asahi-gpu
<jannau>
first VK cts on asahi-6.13.8-1 + dts fix + mesa asahi-20250221 finished without GPU fw crash and only a single "** GPU timeout nya~!!!!! **" at the end